Musk shares an AI industry self-deprecating meme:
Back then I proved theories; now I just stack GPUs

Looks like 2028 is still the age of the storage sector

Musk reposted an “AI then / AI now” comparison chart, and only commented with one word: “Accurate ?”
In the image, past machine learning research involved designing kernel methods, proving why models work, and building probabilistic models. In the era of large language models, the process has been reduced to three steps: make the Transformer bigger, feed it more data, and add more GPUs—so the model gets stronger again, and again.

This is the grim reality of the large-model industry: in the past, you studied principles first, then designed models; now, you often scale up first, and then investigate why capabilities emerge.